Help with RTX 3080 Throttling on Complete Hydro X build

[tateconcepts]

I built this TIE Fighter build for ML with VR use and recently been using it for a boost in GPU mining. I have an issue where I am experiencing what appears to be coolant evaporation over time and thermal throttling on the dual RTX 3080 graphics. All components are cooled by Corsair components with an XD5 pump, dual (2) XG7 RGB 30-Series Ventus GPU water blocks, an XC7 CPU water block, dual (2) XR5 360mm radiators, Three (3) QL120 fans and Three (3) SP120 fans.

Over the past month or so since I filled the reservoir back within 1cm or so, it has dropped again to almost 3cm which is almost a half an inch in just under two months. I don't see any leaks anywhere and tested this with a paper towel in various areas when I brought the build online. It has consistently had warmer than expected temps since it was built, however I feel I should be getting about 10C lower temps on the GPU. I have an external Aorus RTX 3080 Gaming Box about six feet from that unit and its running at 40-43C. I know the soft tubing is not ideal however I felt it best until the temps can be optimized.

The current flow of this is XD5 Pump > XC7 on i9-10850K > XR5 Radiator (top exhaust)  > XG7 on RTX 3080 > XG7 on RTX 3080 > XR5 Radiator (side intake)

Feedback and thoughts welcome, thank you!

Edited November 16, 2021 by tateconcepts

[tateconcepts]

Correction, the gaming box is running almost opposite at almost 58C but these RTX are still 49-50C.

[LeDoyen]

50°C while mining on both GPUs?

If that's the case it's fairly cool, given that you're dumping something like 700W of heat into two 360 rads.

To lower the temps you could flip the side intake fans as side exhaust, and also flip your back fan to intake. That should help, not dumping the heat from the front rad into the top one. In your usual rig with only one GPU, it helps by a couple of degrees, sometimes a little more. but with the amount of heat these two must output the change may be more drastic.

Worth a shot, it should be easy if the fans are set as pull, easy  access

[tateconcepts]

Thank you! It is indeed 50C or so for both. TDP for all is around @550W actually (says the smart plug). I think the Corsair HXi has some additional details on that too.

Hmm, I don't have a rear fan and the top is pushing all that heat from initial at the CPU to have all of the residual heat also pass through it. Just adding a fan and making the top another intake might be a good idea!

Any thoughts though on the coolant loss? I'm like just over an inch now and I just filled this almost to the top about 5-6 weeks ago I think. This feels somewhat excessive for water cooling?

[LeDoyen]

you may want to look at the AXi power reading yes, or directly the GPU power draw in the icue dashboard.

if they run at 50°C with 340W each, that's decent. If that's the temperature for 200-220W each as the smart plug implies, then it's not impressive for watercooling, but they are still far from throttling (they stop boosting around 83°C and drop clocks beyond that).

As for your water loss yes it's not normal at all. if there's no dripping anywhere, make sure the reservoir is properly sealed.

It's also possible you still have air bubbles in the loop that works its way out gradually, and you'd see that as level drop.

When you started the loop for the first time, did you tilt the case with the pump at full speed to bleed all the air out? can you hear some gurgling when you set the pump at 100%?

Also out of curiosity, what is the coolant temperature when the GPUs are at full load?

[tateconcepts]

So it looks like I'm running both indeed around 210-220w at just over 50C but they are throttling because of thermals and I can see that in GPU-Z. These GPU are at full load all the time as they are mining so what you see is that temp you are asking about.

I don't know I did any tilting when I first built this six months ago but it's been running at at full GPU load almost that entire time, either training models or GPU mining otherwise. During that time,. it has already lost 3/4 of the coolant before I refilled it therefore I would presume those air bubbles have long left the system. I think this is sealed but clearly it has to be leaking air somewhere. By the looks of the reservoir, it's almost like the condensation indicates the coolant is not getting cooled enough and possibly leaving there or through the ports on the top somehow?

 

Edited November 22, 2021 by tateconcepts

[tateconcepts]

It might help to actually see the build. Also, I do believe I still had a few improvements to do. I wasn't amazed by the LL120 fan performance but they came with the Hydro X kit. Frankly, I wish they didn't but that's what I got and used those for the intake you see in the image. The top radiator however, I opted to use the two fans that came with the 5000D case and added one Corsair Black 120mm High Performance Airflow Desktop fan. The performance on that fan however is not so good and I speculate those that came with the 5000D were AF designs, which I could probably get better cooling by changing those.

 

[Mark_T]

Personally I'd be running both sets of fans as exhaust as recommended by Corsair.

You have an airflow case there which is designed for negative pressure and you really want to get the hot air straight out of the system so the cooler air will be drawn into the case.

What is the ambient temperature of the room?

[LeDoyen]

ouch.. those clocks at full load point at a probable mounting issue on the waterblocks. it's throttling hard despite the low power limit you've set.

it's possible part of the die isn't touching the coldplate at all. it shouldn't drop clocks at 50°C.

When mounting the waterblocks, to compress the thermal pads on the VRAM, you have to tighten the screws fully. Sometimes people stop tightening because it becomes harder, but you have to tighten them till it stops, or else the coldplate doesn't touch the die. Could be the issue.

Maybe before taking the build apart i'd check the GPU on HWinfo64, to look at GPU memory junction temperature.

I am only used to gaming loads, and what you do seems to load the memory a lot more than the GPU itself. I wonder if the thermal limit may be hit on memory temp rather than GPU die.

HWinfo will show you GPU temp, GPU memory junction temp, and Hot spot temp.

Also, regarding the water issue, do you have the water temp sensor plugged somewhere?

[Mark_T]

37 minutes ago, LeDoyen said:

Maybe before taking the build apart i'd check the GPU on HWinfo64, to look at GPU memory junction temperature.

I am only used to gaming loads, and what you do seems to load the memory a lot more than the GPU itself. I wonder if the thermal limit may be hit on memory temp rather than GPU die.

Agreed, I have a couple of 3080's mining with only air cooling and those GPU temperatures although high for water are not high enough to start throttling anything, the memory will be a lot higher though and far more likely to be the source of the problem, which is another reason why I'd want to be exhausting heat from the case as much as possible...

[tateconcepts]

It's possible that the temp sensor is not plugged in, or at least I don't see it anyways.

In regards to @Mark_T but it is throttling and we can see that in the GPU-Z image I provided. Thanks for the tips everyone, I'm going to check for the temp sensor next and then maybe change the fans or add another and see what happens. Considering it's been this way for six months, it will be awhile before I probably wish to drain this again. I'll update once I have these done.

 

Thanks everyone for the comments!

 

[LeDoyen]

the temp sensor is supposed to be a G1/4 plug screwed into the reservoir with a cable attached, and plugged to  either a corsair controller, or to the thermal sensor input of  your motherboard if you control fans and pump from it.

I believe it comes with the XD5, so you may have left it aside at installation perhaps? If it's not installed by all means do it 🙂 it's possible your water gets way too hot which may explain that loss of coolant to an extent.

And you should control the fans speed based on coolant temp too so you really need that sensor enabled.

[Mark_T]

2 hours ago, tateconcepts said:

In regards to @Mark_T but it is throttling and we can see that in the GPU-Z image I provide

Yes, it is throttling, but I suspect it is the memory junction temperature you need to be checking, not the GPU core temperature.

Get the latest version of GPU-Z and look at the 'Memory Temperature' sensor...

 

This is my 3080FE mining with a few foreground activities at the moment...

[Corsair HP]

On 11/24/2021 at 2:50 PM, tateconcepts said:

It's possible that the temp sensor is not plugged in, or at least I don't see it anyways.

In regards to @Mark_T but it is throttling and we can see that in the GPU-Z image I provided. Thanks for the tips everyone, I'm going to check for the temp sensor next and then maybe change the fans or add another and see what happens. Considering it's been this way for six months, it will be awhile before I probably wish to drain this again. I'll update once I have these done.

 

Thanks everyone for the comments!

 

Please download HWinfo 7.13 or later - this version should give you the option to see GPU Hot Spot Temperature - check this before doing anything else.

If this sensor shows 70-90C when mining, then you most likely have bad contact on your block.